How Crypto Can Survive If AI or Quantum Computers Break Its Cryptography

Public-key cryptography rests on mathematical problems believed to be hard, but their hardness is not proven. The speakers agreed that quantum computers running Shor’s algorithm are expected to break elliptic-curve cryptography, though when such machines will be capable is uncertain. Whether AI could also break public-key cryptography remains unknown. Justin Drake believes AI may uncover failures humans have missed; Yehuda Lindell said there is no evidence that the decades-old assumptions behind elliptic curves are failing and called claims of an AI threat unsupported by logic. The speakers cautioned that no public evidence cannot rule out undisclosed discoveries.

ECDSA signatures authenticate control of an account and authorize blockchain actions. Their small size and efficiency on constrained devices help explain why Bitcoin and Ethereum have used elliptic curves since their beginnings; Connor Deegan said about 99% of blockchains use them for transaction signatures. A sudden break could therefore put exposed keys and assets at risk. Dan Crowdfiest warned that if attackers could recover keys before white-hat hackers secured assets, coins could go to zero. Unlike banks or TLS certificate authorities, blockchains have few safeguards between signing a message and transferring funds, and may be difficult and slow to upgrade.

The speakers discussed lattice-based and hash-based schemes as possible replacements, but neither is established as the universal answer. AI attacks on Hawk, a lattice-based scheme, and Macalese do not show that current NIST standards are broken: neither scheme was used in production, and the attacks exploited different weaknesses. Drake favors hash-based cryptography, arguing that elliptic-curve and lattice systems have mathematical structure AI might exploit, while hashes are designed to behave more like random machines. Eli Ben-Sasson questioned reliance on lattices, including succinct module SIS versions dating only to 2012. NIST has standardized both a lattice-based and a hash-based post-quantum scheme, with lattice-based ML-DSA its primary recommendation.

Each option involves trade-offs. Hash-based signatures are larger and slower, and do not provide a hash-based equivalent for sharing a secret. Most blockchains do not use encryption, but systems such as Zcash use encrypted notes; breaking that encryption could reveal past activity without necessarily enabling theft. Hash functions are not perfectly random, and speakers noted that they have been retired or broken more often than some other cryptographic constructions over the past 30–40 years, though that comparison is imperfect. One speaker described heavily used SHA- or Blake-family hashes as resilient, while acknowledging that no one knows whether elliptic curves or hashes are more vulnerable to AI.

Quantum risk is more predictable in timing than AI risk, but remains a long-term engineering challenge. Alex Pruden put the earliest realistic quantum attack around 2030, with an attack more likely than not in the early to mid-2030s; he would be surprised if attack-capable quantum computers existed by the end of this decade. Google’s magic-states result and a Quantinuum error-correction result showed engineering progress but did not change that assessment. Drake raised a more immediate worst case for AI: ECDSA could plausibly be broken before “Q-Day,” perhaps within months rather than years, defining a break as recovering private keys quickly, such as within a week using a large GPU cluster. No production scheme has been broken by AI so far.

The speakers urged preparation, not panic: Bitcoin and Ethereum still use elliptic curves, and one speaker said they would not migrate for at least two years. A post-quantum migration using hash-based signatures could also address potential AI risks, but choosing and implementing a scheme requires care. Crypto agility—building systems with backup options and the capacity to change cryptography—would reduce dependence on any one choice, especially over the 30-to-50-to-100-year security horizon Drake highlighted. The speakers said migration planning should start soon, while noting that the first post-quantum schemes adopted by Bitcoin or Ethereum will not be the last.

Emergency proposals included Haseeb Qureshi’s ecosystem-wide plan for blockchains, wallets, exchanges, and apps: establish a hash-based recovery mode, encourage users to create backup keys, and eventually block transactions for those without them. He also proposed a pre-deployed validator kill switch that could be activated without a protocol upgrade. Balaji Srinivasan suggested Ethereum users pre-consent on-chain to rewinding state by a set period, such as a week, and reversing transactions after a signature or hash function is hacked. The speakers stressed that consensus is the central obstacle, and that each chain must consider its own recovery, rollback, and redundancy policies. A fork or restart could be possible, but a sudden theft would still be profoundly disruptive.